TRANSITION-LEVEL EQUIVALENCE AUDIT: TWO CLOSE IMPLEMENTATIONS 3 October 2026 UTC. Partial audit; does not clear the full novelty gate. Question Can a released fixture distinguish the paired pending-operation case in which current verification takes time, cancellation becomes effective after its acknowledgement, replacement can miss a rescue deadline, and stopping leaves background commitment active? Compare transitions and endpoints, not labels. Local reference benchmark/necessity_pending.py implements submission before tick advancement, delivery before physical events, configurable cancellation/commit tie order, irreversible expiry, delayed replacement, and settlement after stopping. reviews/necessity-timing-boundary.json records48 verified fixed-policy traces. These are public development controls. They do not establish a new metric, an all-policy feasibility frontier or independently authored confirmation cases. ADMITBench Pinned revision8b859de77334872efc3fab1c5a764b43d941f210. Re-fetched selected physics and procedure files; byte hashes match the earlier inventory. CSTR C07 accepts protective action and rejects escalation when human response exceeds its hazard window. C12 accepts escalation or verification under sparse evidence. The physics implementation projects consequences of an advisory action. This directly precedes harmful procedural delay. These two fixtures do not supply our paired pending-cancellation timeline; that observation is bounded to them, not a claim that the full framework cannot implement it. Our timely-intervention family should be presented as an extension/control, not a first demonstration. Evidence: reviews/necessity-prior-art/admit-bench-inventory.json, C07/C12 records, physics.py World.project lines114-148 and recovery lines150-173. ConsequenceBench Pinned revision469a4effe7de953bab56c9fc43853131541af901. Re-fetched selected pressure_episode.py and control_planes.py; byte hashes match the inventory. The pressure executor advances the clock per tool, materializes dynamic records, and stores proposed effects, readbacks and compensation state. Evaluation combines required inspection, evidence binding, disposition and obligations. In the inspected perform/evaluate methods, no pending-cancel queue or expiring rescue is advanced; compensation sets a flag. The proposed cancellation witness requires additional transitions there, rather than different wording. The specification layer separately names races and delayed duties; do not generalize this executor finding to all released families. It already measures wrong blocking and unsafe effects. Evidence: reviews/necessity-prior-art/consequencebench-inventory.json, pressure_episode.py agent_view838-895, perform897-976, evaluate978-1028, reference_execute1030-1070. No external code executed. Case mapping Feature ADMIT C07/C12 Pressure executor Consequential time Hazard window Clock/dynamic evidence Evidence-dependent permission Yes Yes Harmful non-action C07 unsafe escalation False blocking Pending delayed cancellation Not in these fixtures Not in inspected methods Replacement after cancellation Not in these fixtures Not in inspected methods Post-stop physical settlement Not established here Not in inspected methods Information-matched optimum Not established here Reference uses oracle IDs Repair-boundary metric Not established here Not established here Not established means no verified evidence in this bounded inspection; it is not a negative corpus-wide finding. Oracle reference execution may validly calibrate an evaluator, but is not automatically a deployable observation-limited policy. Decision These selected implementations do not invalidate the specific pending-race fixture, but do invalidate broad novelty claims about time, evidence or harmful inaction. A mechanism difference alone does not establish a research contribution. The stronger claim still needs a working option-preservation/repair diagnostic and comparison against Cordon, Outcome Finality, AgentAbstain and the partial- observation assurance theory. A programmable framework may support an extension without having reported the same experiment. Separate representability from published fixture and measured result. Independent review remains unsecured. Access and reproducibility Primary files were fetched by their exact URLs in the existing inventories and checked with SHA256 before reading selected ranges. A browser request initially used an incorrect pressure path and returned404; the inventory's correct pinned URL succeeded. No404 was treated as absence of implementation. A GitHub browser tree view was restricted; direct pinned public files succeeded. No author was contacted and no access terms were accepted. This audit changes no frozen simulator, collector, analysis or manifest. Full collection remains conditional on remaining contribution, access and budget gates.